Proof for review only

RepRap - The Replicating Rapid Prototyper

Journal: Robotica

Manuscript ID: ROB-REG-09-0184

Manuscript Type: Regular Paper

Date Submitted by the 04-Nov-2009 Author:

Complete List of Authors: Bowyer, Adrian; Bath University, Mechanical Engineering Jones, Rhys; Bath University, Mechanical Engineering Haufe, Patrick; University of Applied Science, Biomimetics/Biological Materials Sells, Edward; Bath University, Mechanical Engineering Iravani, Pejman; Bath University, Mechanical Engineering Olliver, Vik; Diamond Age Solutions Ltd. Palmer, Chris

self-replicating machine, rapid prototyping, additive fabrication, Keywords: biomimetics, mutualism, open-source, free software, fused-filament fabrication

Page 1 of 34 Proof for review only

RepRap – The Replicating Rapid Prototyper

Rhys Jones1, Patrick Haufe2, Ed Sells1, Pejman Iravani1, Vik Olliver3, Chris Palmer4, and Adrian Bowyer5

1Mechanical Engineering Department, Bath University, Bath BA2 7AY, UK. 2Faculty 5, Department of Biomimetics/Biological Materials, Hochschule Bremen, University of Applied Science, Neustadtswall 30, D-28199 Bremen, Germany. 3Diamond Age Solutions Ltd., 72 Warner Park Avenue, Laingholm, Waitakere, New Zealand. 43 Boatmans Row, Astley Green, Tyldesley, Manchester M29 7JQ, UK. 5Corresponding author: [email protected]; Mechanical Engineering Department, Bath University, Bath BA2 7AY, UK.

Abstract

This paper presents the results to date of the RepRap project – an ongoing project that has made and distributed freely a replicating rapid prototyper. We give the background reasoning that led to the invention of the machine, the selection of the processes that we and others have used to implement it, the designs of key parts of the machine and how they have evolved from their initial concepts and experiments, and estimates of the machine’s reproductive success out in the world up to the time of writing (about 2,500 machines in a year and a half).

Keywords: self-replicating machine, rapid prototyping, additive fabrication, biomimetics, mutualism, open-source, free software, fused-filament fabrication. Proof for review only Page 2 of 34

1. Introduction

RepRap is an open-source self-replicating rapid prototyping machine. It is a robot that uses fused-filament fabrication1 to make engineering components and other products from a variety of thermoplastic polymers. RepRap has been designed to be able automatically to print out a significant fraction of its own parts. All its remaining parts have been selected to be standard engineering materials and components available cheaply worldwide. As the machine is free2 and open-source anyone may – without royalty payments – make any number of copies of it ether for themselves or for others, using RepRap machines themselves to reproduce those copies.

In this paper we briefly examine the terminology and history of artificial reproduction, and then describe the biomimetic genesis of the RepRap machine, its original design, how and why that design has changed into its current form, RepRap’s global adoption and use, and the commercial offshoots and spinouts from it.

1.1 Terminology

Historically, the terminology used in the field of self-reproducing machines has sometimes been unclear, with different meanings being ascribed to the same terms. In an attempt to bring some systematisation to this, we will define key terms for use in this paper at least.

Kinematic machine - A physical machine that is composed of fixed and moveable parts. This term makes a distinction between real machines and software models (which are frequently used for simulation). In what follows we take the words kinematic machine to include living organisms.

1Fused-filament Fabrication (FFF) is sometimes called Fused Deposition Modelling (FDM). But this latter phrase is trademarked by Stratasys Inc., and so is not in unconstrained use. FFF was coined by the members of the RepRap project to give a synonymous term that can be used by anyone without restriction. 2In this paper we always use the word “free” to convey both the meanings that it has in the free software discourse: “Free as in freedom”, and “free as in beer”. Page 3 of 34 Proof for review only

Self-replication – we start with the idea that self-replication could mean an imaginary Platonic process by which a kinematic machine was able to create an exact copy of itself. The Second Law of Thermodynamics and Shannon’s theorem [1] show that information cannot be copied without loss or error indefinitely, implying that the idea of an exact replicator is an impossibility. (It is the errors, of course, that drive Darwinian evolution.) Whilst it is philosophically and poetically useful to have words for impossible ideas, here we reduce the strength of the word replication to give it an engineering meaning: a copy within specified tolerances that will work as well as the original.

Self-reproduction - a process by which a kinematic machine is able to create an approximate copy of itself, perhaps with either insignificant or significant errors. All living organisms are self-reproducers. The specified-tolerances-and-works-as-well distinction between replication and reproduction follows through the definitions below, and the rest of this paper. Replicators are a subset of reproducers.

Self-manufacturing – the ability of a kinematic machine to make some or all of its own parts from raw materials. This clearly prompts a requirement for a definition of “raw”: is an etched PCB a raw material? Or a uniform copper-clad board? Or some copper, some glass, and some epoxy resin? Forensically, many Gordian Knots of this sort are cut by asking, “Would a reasonable person say it is so?” and leaving it at that. We adopt the same approach as the law.

Self-assembly - refers to the ability of a kinematic machine to manipulate a series of parts into an assembled copy of itself.

Autotrophic self-reproduction or -replication -- the ability of a system to make a direct copy of itself from raw materials without assistance. As yet, no artificial autotrophic self-reproducing kinematic machine has been made. However, examples exist in biology (see Section 2). For a kinematic machine to achieve autotrophic self- reproduction, it must contain a number of critical subsystems. One attempt to identify these subsystems was undertaken in Freitas and Merkle’s “Map of the Kinematic Replicator Design Space” in their comprehensive book [2], which identified 137 design properties in order for autotrophic self-reproduction to be possible. Proof for review only Page 4 of 34

Assisted self-reproduction or -replication - a kinematic machine that includes at least one, but not all of the critical subsystems required for autotrophic self-reproduction or replication and so needs human (or other) intervention to reproduce.

By these definitions, RepRap is a kinematic assisted self-replicating self- manufacturing machine, as we shall show below.

1.2 Artificial Reproduction

The concept of self-reproducing kinematic machines has intrigued some of the world’s greatest minds for generations. However, the first person to formalize thoughts on the subject was John von Neumann in the middle of the last century [3]. Much of von Neumann’s work concentrated on his cellular machine, a theoretical and mathematical model, and records of his research into a kinematic (physical) self- reproducing machine are scarce and often informal. Much of the outline presented here is based on the summary in the review by Freitas and Merkle [2].

Figure 1. Schematic of von Neumann’s kinematic reproducer from [4]

Von Neumann’s kinematic reproducer, as illustrated by Cairns-Smith [4] in Figure 1, consists of five distinct components, namely a chassis (c), a set of instructions (I), some form of machinery (m), a controller (r) and finally a sequencer (s). Page 5 of 34 Proof for review only

In order for the kinematic reproducer to function properly it is required that it resides in a stockroom containing an unlimited quantity of spare parts. The kinematic machine features a mechanical appendage, which is able to gather parts at random from this stockroom; the randomly-selected part is inspected and compared to the kinematic machine's instructions. In the event the part is not required, it is replaced in the stock room and the process is repeated until a required part is found. This process is then repeated to find the next required part, and the two parts are connected together using the mechanical appendage. This cycle continues until a physical copy of the kinematic machine is produced, at which point the instructions are copied to memory in the child kinematic machine before it is finally activated.

In devising his kinematic reproducer, von Neumann ignored any fuel and energy requirements. Even so, with a part-count of the chassis estimated at 32,000, the feasibility of the device was poor. Nevertheless the concept did at least demonstrate the principle of a self-reproducer, and has inspired many more people to research further. Most of this work may be broken down into three distinct subsets. Using the definitions above these are:

1.2.1 Autotrophic self-reproducers

With limited success so far into the areas of self-assembly and self-manufacturing, an artificial autotrophic self-reproducer remains an un-achieved utopia for the subject. Whilst theoretical work has been undertaken into the area, all concepts presented thus far are extremely vague on the engineering involved in artificial reproduction, being described by Dyson thus: “We don’t have the science yet; we don't have the technology” [5].

1.2.2 Self-assembling kinematic machines

Some of the most elegant work into self-assembling kinematic machines using special pre-made parts was conducted by Roger and Lionel Penrose in designing their so- called block reproducers [6]. Perhaps the biggest achievements of their design are its neatness and simplicity. Proof for review only Page 6 of 34

The block reproducer (Figure 2) consists of a series of wooden blocks which are placed on an agitating surface. The design of the blocks is such that an interlocking profile exists on each block. "Brownian-motion" is induced into the parts by agitating the surface, enabling the locking profile to be utilised to complete the assembly process. They also designed a more complicated two-dimensional reproducing kinematic machine along similar lines.

Figure 2. A 1-D self reproducing kinematic machine made from parts of two kinds from [6].

Further work into self-assembling processes was conducted by Moses [7], who developed a self-assembling kinematic machine in the form of a Cartesian manipulator based on sixteen types of snap-fit parts. In a similar way to von Neumann’s kinematic reproducer, if supplied with sufficient parts, it was able to build a copy of itself. However, whilst the concept proved promising, the structure of the design lacked stiffness, leading to the machine's requiring external assistance to complete the reproduction cycle. But inspired by this success, the world’s first semi- autonomous limited-part self-assembly kinematic machine was created in 2003 by Suthakorn et al. [8], with an assembly time of just 135 seconds. It consisted of an original robot, subsystems of three assembly stations and a set of subsystems from which replicas of the original robot were assembled.

1.2.3 Self-manufacturing kinematic machines

To date, the amount of research into self-manufacturing process has been limited. The main research into this area was conducted by the Reproducing Concepts Team at Page 7 of 34 Proof for review only

NASA [9] in 1980-1982. They reveal two fundamental models for a self- manufacturing process:

1. The unit growth or factory model consists of a series of sub-assemblies which collectively are able to manufacture and assemble all sub-assemblies within the model until the surrounding resources are exhausted. As the name suggests, and as observed by von Neumann, any machine shop with sufficient equipment may be considered a self-manufacturing unit growth system.

2. Unlike the unit growth model, the unit replication3 model specifies that one device must be able to manufacture all of its own parts. Perhaps the most interesting facet of this model is that it potentially has the ability to be substantially more compact than the unit growth model, to the degree that such a kinematic machine could exist in every home. As yet, an autotrophic unit reproduction model has not been realised. One possible reason for this is that traditionally manufacturing methods require tools to have one specific function, such as a lathe for cutting about an axis, severely limiting the potential designs that can be manufactured with a single kinematic machine. Therefore, the goal of achieving a self-manufacturing process based on the unit reproduction model cannot be realised until an extremely versatile manufacturing technology is realised.

2. The genesis of RepRap

Sometimes the train of thought that instigated a project can become obscured by the progress and the reporting of the project itself. Typically, that train of thought was incomplete, or sometimes downright wrong. In this section we attempt to set down as a matter of record the ideas – good, bad, and indifferent - that led one of us (AB) to invent RepRap. Of course, even this setting down tends to tidy those ideas up (a true record would be so disjointed and inchoate that it would be unreadable), but what follows is the essence of the matter as best we can legibly report.

3We shall call this the unit reproduction model from here on, in accordance with our defined terminology. Proof for review only Page 8 of 34

Understandably, the design of most practical artificial reproducers starts with proposed solutions to the many technical problems of getting a kinematic machine to copy itself. But RepRap was not instigated in that way at all. RepRap was instigated by biomimetically considering extant natural evolved strategies for reproduction.

Biologists categorise most bacteria, archaea, protists, and plants as autotrophic because they are capable of self-nourishment using inorganic materials as a source of nutrients and using photosynthesis or chemosynthesis as a source of energy. However, almost without exception, all the natural species of reproducers in the world (including those in the previous sentence) depend upon other species in some way for their survival and successful breeding – by this light they are all assisted self- reproducers. A few lithophile micro-organisms can survive alone in what are essentially mineral environments, but their numbers are vanishingly insignificant compared with those of the interdependent species. Clearly primordial organisms must have been completely autotrophic, but now that way of life has all but disappeared because the environment in which modern organisms evolved consists, to a first approximation, entirely of other reproducers.

And yet research into artificial reproduction often concentrates upon making the reproducer as autotrophic as possible (like the lithophiles), and researchers regard this as an important aim. Clearly this aim is important for an extraterrestrial reproducer, but why also for a terrestrial one? Why try to follow a strategy that biology has all but abandoned? An artificial reproducer designed to be interdependent with the natural reproducers that will make up its environment would be more likely to be successful.

Dependencies between species take one of three forms: predation, commensalism, or mutualism. Predation is well-understood: lions eat antelope; antelope eat grass. Commensalism usually implies some sort of scavenging -- lions and antelope are uninterested (though not ultimately disinterested) in what the grass does with their

4 dung, their urine, and their exhaled CO2. Mutualism implies a symmetry of

4 Sometimes called symbiosis, but that term (which can mean any species-pair relationship including that between predator and prey) is being replaced by the more precise mutualism for mutually beneficial relationships. Page 9 of 34 Proof for review only

dependencies giving benefit to both partners: the pistol shrimp digs a burrow in which both it and a goby fish live; the shrimp is nearly blind and the fish warns it of danger.

This variety of dependencies prompts a choice in the design of an artificial reproducer: which type of dependencies should our artificial reproducer exploit, and with which natural species? Beneficial options to people might include predation upon pests, commensal gathering of waste, or mutualism with a species the welfare of which we wished to promote (such as an endangered or an agricultural one).

But clearly the most interesting natural species with which our proposed artificial reproducer might exhibit a dependency is ourselves. This makes the choice more sharply cut: it would be foolhardy to make ourselves the prey for our artificial reproducer; having it collect our waste commensally might be useful; but the option most pleasing to our evolved senses of morality and symmetry would be to make ourselves a reproducing mutualist. In other words, we should make an artificial reproducer that would benefit from us, and we from it.

The most famous mutualism in nature, and the one that we all learn about first at school, is a payment for services. Samuel Butler said of it in Erewhon [10] 5:

"Does any one say that the red clover has no reproductive system because the humble bee (and the humble bee only) must aid and abet it before it can reproduce? No one. The humble bee is a part of the reproductive system of the clover."

And, he might have added, the bee is paid in nectar.

The mutualism between the flowers and the insects evolved about 140 mya in the late Jurassic and is one of the most enduring in all biology. For both sets of species it is an evolutionarily-stable strategy corresponding to a particularly unshakable Nash equilibrium.

5 Famously, this novel is also one of the first places in history where the idea of an artificial reproducing machine appears. Proof for review only Page 10 of 34

What service could our mutualist reproducer ask of us? And with what could it reward us?

It would seem sensible to play to the differing strengths of artificial kinematic machines and of people. Artificial kinematic machines can make objects accurately, repeatably and tirelessly. In contrast, they fumble at manipulative tasks that would not tax a small child. People are exquisitely dexterous. (Aristotle called the human hand, "the instrument of instruments.") But – though with practice people may carve and mould beautifully – they cannot do so accurately, repeatably and tirelessly.

So our self-reproducing kinematic machine could be designed to manufacture a kit of parts for a copy of itself, and to need the assistance of people to assemble that copy (that is, it would be an assisted reproducer along the lines of NASA's unit-reproducer [9]). The people would be the humble bee, and the kinematic machine the clover. And what of the nectar? If the kinematic machine were sufficiently versatile to make its own parts, the chances are that it would also be able to make many other items useful to people as well. When it was not reproducing itself, it would be rewarding its assistants with a supply of consumer goods. This idea of a self-reproducing machine also making useful things for people is not new. It goes back through von Neumann to Butler. But we contend that regarding this as a form of biological mutualism and deliberately seeking to achieve that in order to position both reproducers at an evolutionary Nash equilibrium for each is a novel idea.

This was the genesis of the RepRap machine. It was designed to make its own parts to be assembled by people into another RepRap. The people would be driven to do this by the fact that the machine, when not reproducing, could make them all manner of useful products. It seemed (and still seems) likely that this would lead to a mutualist relationship between people and the machine that would inherit some of the longevity and the robustness of the evolutionarily-stable strategies of the bee and the clover.

Finally in this section, we note that clover does not attempt some biological equivalent of copyrighting or patenting the “intellectual property” of its genome. Page 11 of 34 Proof for review only

That genome builds the clover with the sole intent6 of spreading itself with the most promiscuous fecundity possible. Any genome mutation that arose that – for example – attempted to extract some payment (like the nectar) in return for a copy of itself would clearly have a lower reproductive fitness. The nectar and the information are not in any way equivalent. The nectar is a real material resource. It is real property. In contrast, the immaterial genome information has been arranged purely as a result of its success in copying itself as freely as it can, and any impediment placed in the way of that would be to its detriment. For this reason it was decided to follow the principles of the free software movement and to distribute every piece of information required to build RepRap under a software libre licence that requires no royalty payments whatsoever. This would allow private individuals to own the machine, and to use it freely to make copies for their friends.

The RepRap machine is intended to evolve by artificial rather than natural selection; that is, to evolve as the Labrador has from the wolf, rather than as the wolf has from its ancestors. It is hoped that this evolution will come about by RepRap users posting design improvements on-line that may be adopted in future designs of the machine and then downloaded by old and new users in turn. For this reason the GPL was chosen as the RepRap licence, as that obliges people who improve the machine to make public their improvements under a similar free licence [11].

3. The first RepRap machine

In order for the RepRap project to progress, the rather abstract reasoning laid out in the sections above had to give way to some down-to-earth engineering. It is that that occupies the rest of this paper.

6 Of course, no genome has intent. But they all behave as if they do. Proof for review only Page 12 of 34

The first engineering decision was to use rapid prototyping7 as the manufacturing technology for RepRap as opposed to – say – CNC milling. The reasons for this were threefold:

1. Rapid prototyping requires very low forces to create solids, unlike machining, 2. Of all the manufacturing technologies, rapid prototyping is the easiest to control completely automatically by computer, and 3. It is the closest current technology to the “maximum versatility” specified by the 1980s NASA report [9].

The massive cast bases required by machining centres to overcome cutting forces and vibration would be a significant impediment to self reproduction; this was the basis of Point 1. It is true that cold-casting of – for example – concrete might have reduced this impediment to a certain extent, but that would still leave the fact that cutting (especially of complicated re-entrant shapes) requires very sophisticated toolpath- planning algorithms. In contrast, all rapid prototyping requires algorithmically is the ability to compute a sequence of planar slices through a geometric model of the part to be manufactured and to fill each with a hatch or similar pattern (Point 2.). These are straightforward.

At around this stage it was decided (as mentioned in the Introduction) that any parts that the RepRap machine could not make for itself had to be cheaply and widely available to maximise the ease – and hence probability – of reproduction.

Having chosen rapid prototyping, the next decision that needed to be taken was which of the extant processes to use, or whether it would be necessary to invent a new one.

This decision was made by elimination: any process needing a laser was rejected owing to the unlikeliness of being able to use rapid prototyping to make a laser, and the fact that they are expensive items to buy. This removed selective laser sintering

7The term “additive fabrication”, which is now becoming a carry-all for the totality of rapid prototyping and 3D-printing technologies, was, at the start of the RepRap project (2004) less current. Then “rapid prototyping” was by far the more popular term. Its subsequent decline has tracked the changing of use of these technologies from prototyping towards production. Page 13 of 34 Proof for review only

and stereolithography from consideration (and also electron-beam melting for very similar reasons, though that does not actually need a laser).

Similarly, any process needing inkjet print-heads was rejected. Again, it was thought unlikely that the machine would be able to make these for itself, at least initially. Though ink-jet heads are relatively low-cost, they are constrained by the fact that their manufacturers all have a business model of discounting the printers that use the heads and putting a big mark-up on the heads themselves. They sometimes also put chips in the heads to prevent their being re-filled, and adopt other restrictive strategies that make ink-jet an unattractive technology for a machine intended to reproduce as freely as possible. These facts removed ink-jet printing from consideration.

Figure 3. RepRap Version I “Darwin”. This is the first production RepRap machine. Its rapid prototyped parts (white, blue, and green) were made in a Stratasys Dimension commercial RP machine. The cube of the machine has side lengths of about 500 mm. This machine was built in May 2007.

The project was then left with laminated object manufacturing and fused-filament fabrication from the extant technologies. Laminated object manufacturing was attractive because of its simplicity, and because of the ubiquity of its working material: paper. But fused-filament fabrication offered the possibility of being able to build with multiple different materials. This in turn offered the significant advantage in the Proof for review only Page 14 of 34

future of being able to have the machine make a larger proportion of its own components than could be created out of just one material. This, combined with the fact that it was conjectured that fused-filament fabrication [12] could be implemented using low-cost garden-shed methods8, led that to be chosen for RepRap. Thus it was not necessary to devise a new rapid prototyping technology.

Figure 3 shows the first production RepRap machine. There were experimental machines made before this to try out various ideas, but this is the first model used to make copies of its own rapid prototyped parts. It is a stepping-motor-driven Cartesian robot consisting of an open frame made from M8 threaded steel rods held together by rapid prototyped parts and M5 screws. The build base upon which parts are made is cut from 12mm medium-density fibreboard (MDF). Virtually all the parts that the machine cannot make for itself other than the electronics and the motors can be obtained from an ordinary high-street hardware shop. The horizontal X and Y axes (which need to move comparatively fast – typically with a feed rate of 3,000 mm/minute) are driven directly by toothed timing belts. The Z axis only moves by a small distance when one layer of production is finished and the next is about to start; aside from that and the need to return to its home position at the start of a new build, the Z axis is quiescent. It was thus decided to use a screw drive for that. The MDF build plate has eight M8 nuts attached at its corners which are driven up and down by four M8 threaded bars synchronised by another timing belt. The nuts are in pairs held apart by springs to eliminate backlash.

The entire machine is designed to work off a single 12-volt power supply. This can be cheaply obtained by using the power supply from an old PC. It also means that the machine would work off a car battery where no mains electricity was available. The machine consumes about 60 W when running.

Given the choice of fused-filament fabrication a key – indeed the key – part of the machine is its polymer extrude head. This is described in the next section.

8A conjecture that the project has subsequently proved correct. Page 15 of 34 Proof for review only

3.1 Extruder design

Before a fused-filament fabrication extruder was designed, a polymer had to be chosen for it to extrude. It was decided to use polycaprolactone (PCL) initially because:

1. It has a very low melting point (about 60o C), and 2. It is strong (comparable to nylon).

As will be seen below, this polymer was subsequently abandoned for several reasons. But at the start the low melting point (implying ease of heating) was thought to be a critical factor.

A fused-filament fabrication extruder has two main sub-assemblies:

1. The transport; this forces a filament of the polymer into... 2. ...the melt chamber and nozzle.

It was decided to standardise on a 3mm-diameter9 filament as a feedstock because this is a dimension commonly chosen for plastic welding rod, which is very widely available.

3.1.1 Polymer transport

Figure 4 shows the first polymer transport mechanism designed for RepRap. A 12- volt DC geared electric motor (A) drives a stack of pinch wheels (C) through gears (behind the device). The pinch wheels are actually the heads of M4 cap screws; the knurling on the outside of these gave grip. The threaded rods at the front allow the pinch-wheels to be moved together, increasing the pinch force. The number of pinch wheels in the stack could be varied. The 3mm polymer filament (B) was driven down

91/8” diameter in the USA. 1/8” - 3mm = 0.175 mm, a difference easily accommodated by all the extruder designs RepRap has used. Proof for review only Page 16 of 34

into a melt chamber and nozzle at D (see below). The control electronics are at E. All the white parts in the picture except D were rapid prototyped.

Figure 4. The first RepRap polymer transport mechanism.

This device worked, but it was heavy and complicated. The large number of pinch wheels in the stack (4) was needed because of the very low friction coefficient between the pinch wheels and the PCL (despite the knurling).

Shortly after this device was made one of us (VO) came up with a much simpler design. Figure 5 shows it. A threaded rod is forced against the 3mm polymer filament, which runs in a channel. As the thread turns it forces the polymer downwards. This design, which has only one moving part, gives exceptional grip against even the slipperiest polymers, and has a very high mechanical advantage, giving a large potential extruding force. Page 17 of 34 Proof for review only

Figure 5. The screw-driven RepRap polymer transport (inventor's first sketch).

The design of the screw-driven extruder went through several iterations, finally ending up as shown in Figure 6.

The geared drive (at the top) is offset, with its torque being transmitted by a flexible coupling (a short length of steel hawser) – the grey curve just to the right of the white polymer filament. This arrangement allows a straight run of filament down the device, which it was thought might be useful for brittle or stiff materials. However, this straight run was never, in fact, used (the polymer was always flexible enough to feed in at an angle). Further, the flexible drive was a weak point in the design, as it tended to fatigue after being used for about fifty hours. The springs at the back of the device set the force between the screw thread and the polymer, and allowed some compliance as slight changes in polymer diameter moved through the device.

As these developments were taking place, the temperatures that could be easily achieved in the melt chamber (see below) were rising because of design improvements in that. This meant that it was possible to abandon PCL, which didn't just give problems because of its low friction, but was also very sticky and stringy as it was being extruded. These shortcomings led to low-quality built parts. Proof for review only Page 18 of 34

Acrylonitrile butadiene styrene (ABS) was adopted instead. This gave much better build quality (as it was more paste-like upon extrusion, and less viscous). ABS also allowed a return to a much simpler pinch-wheel transport mechanism as, being harder and exhibiting higher friction, it was easier to grip firmly.

Figure 6. The last screw-driven RepRap polymer transport. All the green parts are rapid prototyped.

Figure 7 shows the latest extruder design used in RepRap Version II “Mendel” (see below). A NEMA 17 stepping motor with a knurled shaft pinches the filament against a ball-race. The motor has a 5mm-diameter shaft. With this, a motor torque rating of 0.13 Nm is quite adequate to drive the filament with enough force for reliable extrusion. The device has just a single rapid-prototyped part (green). The stepping motor allows exquisitely precise metering and control of the extruder flow. The four screws holding the motor are in slots. This allows the gap between the motor's shaft and the ball race to be adjusted easily. For 3mm hard-polymer filaments, a 2.5 mm gap works well. This is easily set by putting the shank of a 2.5 mm drill-bit in the device, sliding the motor so that the bit is just trapped between the motor's shaft and the ball-race, tightening the four screws, and then withdrawing the drill-bit. Page 19 of 34 Proof for review only

Figure 7. The current extruder transport mechanism (with melt chamber and nozzle at the bottom). 1: back view of the interior; 2: assembled.

3.1.2 Melt chamber and extrusion nozzle

The requirements of the melt chamber and nozzle were that they should:

1. Be cheap and easy to make, 2. Be compact, 3. Work reliably after repeated heat-up and cool-down cycles, and 4. Not conduct excessive heat to the rest of the machine.

The last point was particularly important, as the rest of the machine would be made from the polymer that the melt chamber would be melting.

Figure 8 shows the first design of melt chamber and nozzle. The 3mm filament enters on the left, and is extruded from an 0.5 mm diameter nozzle on the right.

The white cylinder on the left is a 16mm-diameter polytetrafluoroethylene (PTFE) tube. The internal diameter of the hole running down it is 3.5 mm, which was found worked well with 3mm filament (more on this below). The right-hand end of the PTFE has an M6-threaded hole extending to a depth of 15 mm into which a length of Proof for review only Page 20 of 34

drilled brass M6 studding has been screwed tight. Again, the drilled hole in the studding is 3.5 mm in diameter.

Figure 8. The first melt-chamber and nozzle design. The small squares are 1 mm.

The nozzle at the right end is turned brass. 0.5mm was chosen as the nozzle diameter as it is both about the smallest hole that can be drilled easily, and is also a good compromise between the machine's being able to reproduce fine detail and its not taking too long to fill a large volume.

The heating element (the left-hand pair of wires) was a length of fibreglass-insulated nichrome wire with a resistance of around 10 Ω. This was wound in the grooves of the M6 thread, giving good thermal contact. Temperature sensing was done with a 10 KΩ glass-bead thermistor (right-hand wires). Both these were held on using JB Weld commercial high-temperature epoxy, which is rated up to 315oC.

This design worked, and in particular, because PTFE has a very low thermal conductivity, it kept the rest of the machine cool. But it suffered several shortcomings:

1. The PTFE was held in the polymer transport mechanism by a screw clamp. It tended to slip free of this, because PTFE has such low friction (in contrast, a good thing from the perspective of the polymer filament being forced down the middle of it). Page 21 of 34 Proof for review only

2. 10 Ω was too high a resistance, giving too low a heating power at 12 volts. 3. The JB Weld tended to become friable and was easily damaged after being subjected to a large number of heating and cooling cycles. 4. The 10 KΩ thermistor was not very accurate above about 200oC. 5. PTFE is a rather soft plastic. This meant that sometimes the inner tube swelled under the pressure of the filament being forced into the heated brass tube, leading to an aneurysm of extrudate. The device continued to work while heat was applied, but the swelling could cause blockages when the device had cooled and was then restarted.

The problem of the PTFE slipping in its clamp was solved by turning a series of grooves in the left-hand end of it. It was then epoxied into the transport mechanism with the epoxy keying into the grooves. This held completely firm.

The heater resistance was reduced to 6 Ω on subsequent versions. This gave 24 W of heating, which was ample.

The JB Weld problem was solved by replacing it with fire cement (which is used to seal the flues of central-heating boilers). This is rated up to 1250oC.

The thermistor was replaced by a 100 KΩ one. This corrected the loss of accuracy at high temperatures. Some RepRap users employ thermocouples rather than thermistors. These have the virtue of being linear devices, and so accuracy is not a problem at all. But they do require expensive chips to allow their signals easily to be logged by the analogue-to-digital inputs of microcontrollers (the AD595 and the MAX6675 seem to be the only widespread ones, and each costs about €18). Thermistors are very easy to interface, in contrast. And a look-up table for temperatures makes their use straightforward despite their non-linearity. The mutation from thermistor to thermocouple is an example of one of the many design variations that would not have happened had RepRap not been open-source.

A number of changes were tried to reduce the problem of the PTFE swelling, but this has still not been completely eliminated. Improvements included using Proof for review only Page 22 of 34

polyaryletheretherketone (PEEK) instead of PTFE. This is mechanically much stronger, but the internal friction is higher, and so jams can occur by that mechanism. It is also not such a good thermal insulator, and so it has to be made longer to achieve the same cooling effect. We have recently tried using a PTFE sleeve inside a PEEK outer jacket; experiments with this arrangement are ongoing.

Figure 9. The stainless-steel heat-sink melt-chamber and nozzle.

One of us (CP) has had considerable success by abandoning high-temperature polymers as thermal barriers altogether and instead using stainless steel plus a heat- sink (Figure 9). Here the heating is provided by a 6.8 Ω resistor embedded in an aluminium block heating a copper nozzle. This is connected to the polymer filament transport by a stainless-steel tube cooled by a heat-sink and small fan. As stainless- steel has a low conductivity (for a metal) it is possible to maintain a high temperature differential between the two ends. However, the device can still jam. But this problem can be solved by adding a very slight taper to the hole down the centre of the device from narrow at the top to broader at the bottom. This means that, as soon as the cold polymer begins to move when the device is first started, it detaches from the walls and is subject to almost no friction. This is a very reliable design. Page 23 of 34 Proof for review only

5. RepRap build materials

Figure 10. A RepRap extruder component made in a RepRap machine from ABS. Note the curling at the base.

PCL was quickly abandoned as a building material after initial experiments for the reasons outlined above. In addition, though it is extremely tough, it is also quite flexible. For building the RepRap machine something more rigid – that is with a higher Young's modulus – was desirable. The Young's modulus of PCL is about 1 GPa ; that for ABS is about 3 GPa10. ABS is also very widely available and inexpensive, and so that was chosen next.

Figure 10 shows a typical RepRap build of a reasonably large (80 mm x 60 mm x 30 mm) component of itself using ABS extruded at 240oC. As can be seen, the contraction of the ABS on solidification has caused the bottom of the part to curl away from the base upon which it was built. For smaller parts, this is much less of a problem, but here it is significant. Commercial fused-filament fabrication machines solve this problem by running the entire process in an oven at a temperature of about 70oC. This eliminates curl-up with ABS completely. However, because the point of RepRap is that it should be easy for any technically competent person to construct and

10These values are very dependent on the degree of polymerisation, of course. Proof for review only Page 24 of 34

to run the machine at home in order that its dissemination should be as wide as possible, running in an oven was not really an option. We did experiment with enclosing the build in a roasting bag (intended for joints of meat) and filling it with hot air from a hair dryer. This worked very well, and completely eliminated the problem. But the bag was tricky to set up, it tended to get tangled in the moving mechanism, and a lot of work was required to unload the built parts from within it.

Parallel to the emergence of this curling-up problem, one of us (VO) was experimenting with the use of polylactic acid (PLA) as a RepRap building material. PLA is slightly harder and tougher than ABS. These experiments were not prompted by a need to solve distortion in builds, but by ecological and socio-political considerations. PLA is biologically-sourced and is biodegradable. Its use in RepRap would thus be more environmentally benign than the use of an oil-based polymer such as ABS. Indeed, as it is made from plant matter, building durable objects from it that are kept and not bio-degraded would lock up atmospheric CO2; this would be environmentally positive, as opposed to merely neutral. In addition, it is not too hard for a single individual with access to a small starch crop to make their own PLA from that11. This would mean that such a person with a RepRap machine would not only be able to make the rapid-prototyped parts to reproduce more RepRaps and to make other goods, they would be able to do so with a home-grown polymer supply that was also self-reproducing. And they would be reducing greenhouse gas as they worked.

Serendipitously, it transpires that PLA suffers minimally from contraction problems on cooling, even when builds are conducted using it in a room-temperature environment.

Figure 11 shows a clock made on a RepRap machine from PLA extruded at 190oC. The twelve hour segments were made separately. Each is 140 mm long in its radial direction. With other polymers significant curling would be expected when building an object as long as this. The PLA gives completely flat parts.

11There is one step in the synthesis that is not straightforward: the lactic acid that results from a fermentation of the starch has to be dried to better than one-part in 10 million by weight of water. We have succeeded in doing this by simply passing dry nitrogen over it for half an hour with heating. We conjecture that it would also be possible to do it with air that had been dried by passing it through calcium chloride and then heated. The calcium chloride could in turn be re-used after drying by heating. Page 25 of 34 Proof for review only

Figure 11. A clock made by a RepRap machine from PLA. The face was made from twelve separate segments, each with a longest dimension of 140 mm. (The rather scruffy octagonal backplate is the MDF build platform from an old experimental prototype made to try out ideas for the machine in Figure 3.)

In addition to its positive environmental characteristics and its lack of distortion on building, PLA welds to itself very strongly when one fused-filament fabrication layer is being deposited on top of another. Parts made from it are less susceptible to delamination under stress in the vertical-build direction than are those made from ABS. It is an almost perfect material for the fused-filament fabrication process. Its one disadvantage (aside from the fact that it is only commercially available from a small number of companies) is that it tends to “string” slightly, leaving filaments sticking out from builds where the extrude head has left the build for an in-air movement. But these only take a few seconds to remove by hand with a blade after a build is complete.

6. Reproduction of RepRap machines

Figure 12 shows the first reproduced RepRap machine. The parent machine on the left is the one in Figure 3. All the rapid prototyped parts for the child machine on the Proof for review only Page 26 of 34

right were made in PLA by the machine on the left, except for one grand-child part (a timing-belt tensioner) that the child machine made for itself. That grandchild part was the first part made by the child. It took about twenty minutes to make, and was finished at 14:00 hours UTC on 29 May 2008 at Bath University in the UK.

Figure 12. The first reproduced RepRap machine – parent (left) and child (right).

The child machine was within tolerance of the parent, and worked just as well. RepRap is thus a kinematic assisted self-replicating self-manufacturing machine. The design of the machine includes screw and other adjusters to allow a child machine to be set up to produce parts as accurately as its parent (in just the same way as conventional machine tools are adjusted). RepRap assisted replication is thus not subject to degeneracy.

The fraction of the parts by count of the RepRap machine that it makes for itself is 48% ignoring fasteners (that is, nuts, bolts, and washers – if these are included they are 73% by count of the machine and the RP parts drop to 13%) [13]. It would be straightforward to make a much simpler and rather cheaper machine to an almost identical design by gluing the parts together rather than using nuts and bolts. The machine could easily make small cylindrical locating lugs to fit in the bolt holes and to hold the gluing parts in the correct relative position while the glue set, but that would make it harder to replace parts in the machine and to service it. Page 27 of 34 Proof for review only

Many people have downloaded the RepRap designs from the project website and used them to make RepRaps in commercial rapid prototyping machines. In one case (father and son constructors Bruce and Nick Wattendorf) the design was cut from wood, assembled, and made to work completely successfully.

The RepRap community calls free open-source rapid prototyping machines that cannot make a significant fraction of their own parts, but that can make parts for RepRap machines, RepStraps (from bootstrap). Many private individuals (including one of us – CP) have made RepStraps in order to allow themselves subsequently to make RepRaps. Two companies have been formed to make and to sell RepStraps using laser-cutting rather than rapid prototyping (Bits from Bytes in the UK, and MakerBot Industries LLC in the USA). As required by the GPL, those companies’ RepRap-derived designs are being distributed free and open-source.

Increasingly, people are using their RepRap machines to make sets of RepRap parts for other people, as the project plan intended.

Figure 13. The first RepRap child sale. Left to right: Liav Koren, Michael Bartosik and Wade Bortz. (November 2008)

Figure 13 shows the first complete set of rapid prototyped parts made in one RepRap machine being sold to other would-be RepRap constructors. The parent machine (in the background by the oscilloscope) belongs to Wade Bortz from Canada. The child machine's parts (in the open cardboard box) were bought by Liav Koren and Michael Proof for review only Page 28 of 34

Bartosik of Toronto, who paid one case of Upper Canada Dark Ale for them12. Given the analogy with the payment of nectar for flower reproduction discussed in Section 2 above, the case of beer is particularly appropriate…

Owing to the free distribution of the machine it is difficult to make an estimate of the number of RepRaps and RepStraps that there are in the world. However, the sale of electronics kits for the machine (which are also produced commercially) sets a lower- limit of 2,000 machines. However, some people construct their own electronics rather than buying. 2,500 machines would seem to be a conservative estimate of the total population at the time of writing (October 2009).

Figure 14. Some of the World's RepRap users (courtesy of Google Maps).

The RepRap website invites builders of the machine to mark their location on a map. Figure 14 shows this map as at the end of 2009. Only a fraction of builders have placed themselves on it, but it gives an interesting (if self-selected) sample of the distribution of the machines.

12At the time of writing, the lowest-cost non-open-source rapid prototyping machine (the SD-300 made by Solido Ltd in Israel) was being retailed at about €12,000. Page 29 of 34 Proof for review only

6. Changes made to produce RepRap Version II

The authors and their many RepRap colleagues around the world have just finished the design and commissioning of the latest RepRap machine: RepRap Version II “Mendel” (Figure 15). This incorporates many of the lessons learned from Version I; in particular, lots of improvements and suggestions from the worldwide RepRap community that were posted on-line in the project's forums have been included in the design.

Table 1 gives a comparison between RepRap Version I “Darwin” and RepRap Version II “Mendel”. The cost-of-materials-to-build figures are for a single purchase of all that is required to build one machine from end retailers – it takes no account of bulk discounts and wholesaling.

Figure 15. RepRap Version II “Mendel”. This is smaller, lighter and simpler than Version I, but it has a larger build volume.

The designs of both machines allow their sizes and working volumes to be changed simply by cutting longer or shorter rods to make up the framework, so the values for both of those are nominal. Proof for review only Page 30 of 34

Darwin may be carried a short distance by one person with some difficulty. Mendel, in contrast, can be swung in one hand like a rather bulky briefcase, and is easy to carry anywhere. This has turned out to be a surprisingly important improvement – portability makes it a lot more convenient to work with the machine in many contexts.

RepRap Version I RepRap Version II “Darwin” “Mendel” Cost of materials to build (€) 500 350 Percentage self-manufactured13 48% 46% Size (mm3) 600(W)x520(D)x650(H) 500(W)x400(D)x360(H) Weight (Kg) 14 7 Build envelope (mm3) 200(W)x150(D)x100(H) 200(W)x200(D)x140(H) Deposition rate (ml per hour) 15 15 Positioning Accuracy (mm) 0.1 0.1 Nozzle diameter (mm) 0.5 0.5 Volume of RP parts to build (ml) 1200 1110 Power supply (W) 12 V x 8 A 12 V x 5 A Interface USB/G-Codes USB/G-Codes Host computer Linux, Windows or Mac Linux, Windows or Mac Table 1. A comparison of Darwin and Mendel

The percentage of the machine that it makes for itself has dropped slightly from Darwin to Mendel, but this will rise significantly once the multiple write-head changer and heads are finished (see below).

The deposition rate of 15 ml per hour is the volume extruded by the extruder. In common with commercial fused-filament fabrication machines, RepRap does not usually build parts completely solid – there are some air inclusions. With RepRap the degree to which this happens is completely under the control of the user. It is possible to build parts very fast with a sparse honeycomb interior, or more slowly with a dense interior. Unlike commercial machines, RepRap also allows interiors to be built fully- dense. This reduces build quality slightly, but allows gas- and water-tight parts to be

13 Ignoring nuts, bolts, and washers. Page 31 of 34 Proof for review only

made. With the nominal settings of the machine, the 15 ml per hour deposition rate becomes 19 ml per hour of object built.

As was mentioned above, the 0.5mm nozzle diameter was chosen as a compromise between ease of manufacture, speed of deposition, and fineness of feature resolution. But it is perfectly possible to make other diameter nozzles without changing any other aspect of the RepRap machine's design. Drilling very small holes is difficult, of course. But Jens Kaufmann of Heriot Watt University has suggested it may be possible to make fine nozzles by running copper sulphate solution through an 0.5 mm brass nozzle and electroplating copper onto its inner surface – an experiment that we have yet to try.

We are currently working to add the final detail to the Mendel design: a head changer that will allow the machine to swap extrude heads automatically and so print with multiple materials. The reader will recall that this potential was one of the initial reasons for choosing fused-filament fabrication for RepRap. We also have a paste extruder in the late stages of development (driven by compressed air from a fizzy- drink bottle reservoir charged using a bicycle pump). This should allow RepRap access to the very wide range of materials already proved in fused-filament fabrication by the open-source Fab@Home project [14] (this project, incidentally, was inspired by RepRap).

7. Conclusions

There is not the space here to go into every last reason for adopting certain designs for parts of RepRap and rejecting others, nor to describe all the many alternatives that were experimented upon and not adopted because of the results of those experiments. All this information is, however, available in copious detail on the RepRap blogs, forums and wiki (http://reprap.org).

At the start of 2008 there were four RepRap machines in existence, all made on commercial rapid prototyping machines. A year and a half later we conservatively Proof for review only Page 32 of 34

estimate that there are about 2,500 derived machines all over the world. We have no way of telling how many of those are replicated RepRap machines, and how many are non-replicating RepStraps for making further RepRaps. But judging by the large number of requests for the fused-filament fabricated parts for the new Mendel design which has just been released, lots of people want to make, use, and distribute their own assisted replicator.

One of the members of the RepRap project (Zach Smith) has set up a website where anyone can upload and download free designs for consumer goods to be printed by RepRap and other rapid prototyping machines (http://thingiverse.com). This is a considerable success, with many new designs being added daily.

The authors think (perhaps self-servingly) that RepRap is quite a good artificial reproducer. But even a poor reproducer that is out in the world freely parenting children must improve by Darwinian selection, and so should eventually overtake even the most exquisitely-designed reproducer that stays in the laboratory.

All of human engineering can be considered to be a vast unit-growth reproducer that copies itself with the assistance of – and with benefit to – humanity; a grand version of von Neumann's well-equipped machine shop. RepRap is moving towards compressing as much of that idea as possible into a unit-reproducer that one human may carry in one hand, and may freely copy for their friends.

8. Acknowledgements

We would like to thank the U.K.'s Engineering and Physical Sciences Research Council and Bath University's Innovative Design and Manufacturing Research Centre for the provision of research grants that funded significant parts of this work. We would also like to thank the worldwide RepRap community for their continuous volunteering of ideas, designs, physical prototypes, software, corrections, and suggestions. Without them, the project would be but a shadow of what it has become through their input and support. Page 33 of 34 Proof for review only

9. References

Where works are available freely on-line, URLs are in brackets at the end of the reference.

1. SHANNON, C.E., 1948. A Mathematical Theory of Communication. The Bell Systems Technical Journal,Vol. 27, pp. 379–423, 623–656, July - October, 1948. (http://tinyurl.com/f4two) 2. FREITAS, R.A. and R.C. MERKLE, 2004. Kinematic Self-Replicating Machines. Georgetown, Texas: Landes Bioscience. (http://www.molecularassembler.com/KSRM.htm) 3. VON NEUMANN, J., 1966. Theory of Self-Reproducing Automata. Illinois: University of Illinois Press. 4. CAIRNS-SMITH, A.G., 1971. The Life Puzzle. Toronto: University of Toronto Press. 5. DYSON, F.J. 1972. The World, The Flesh, and the Devil. Self-Reproducing Machinery. The Third J.D. Bernal Lecture, 16 May. London: Birkbeck College. 6. PENROSE, L.S., 1958. Mechanics of self-reproduction. Ann. Human Genetics, 23, pp. 59-72. (http://vx.netlux.org/lib/mlp01.html) 7. MOSES, M., 2002. A Physical Prototype Of A Self-Replicating Universal Constructor. Masters Thesis. Mechanical Engineering, University of New Mexico: New Mexico. (http://home.earthlink.net/~mmoses152/SelfRep.doc) 8. SUTHAKORN, J., A.B. CUSHING, and G.S. CHIRIKJIAN. 2003. An autonomous self-replicating robotic system. Proceedings of the IEEE/ASME Intl. Conf. on Advanced Intelligent Mechatronics (AIM 2003), December Kobe, Japan. (http://tinyurl.com/ylfgr9b) 9. FREITAS, R.A. and W.P. GILBREATH. 1982. Advanced Automation for Space Missions. Proceedings of the NASA Conference Publication CP-2255 (N83-15348), Summer 1980. (http://www.islandone.org/MMSG/aasm/) Proof for review only Page 34 of 34

10. BUTLER, S. 1872. Erewhon, (http://www.gutenberg.org/etext/1906) 11. STALLMAN, R. et al. 1991. The GNU General Public License, Free Software Foundation, Inc. (http://www.gnu.org/licenses/gpl.html) 12. CRUMP S.S. 1989. US Patent 5121329: Apparatus and method for creating three-dimensional objects., assigned to Stratasys Inc. (http://tinyurl.com/ybtrkft) 13. SELLS, E.A. 2009. Towards a self-manufacturing rapid prototyping machine, PhD thesis, University of Bath; (http://tinyurl.com/ygjgva7) 14. LIPSON, H., MALONE, E. et al. Fab@Home; Cornell University and elsewhere (http://fabathome.org)